Top-Down Versus Bottom-Up Approaches to Aspect: The Case of the Dutch Prepositional Progressive

Abstract

Progressive constructions in Germanic are usually studied as progressive constructions—that is, exclusively so. I characterize this as a top-down approach to aspect, which, I argue, harbors the risk of overlooking relevant language-specific structures that are similar in form and meaning. This paper, therefore, advocates taking a bottom-up approach. Based on a case study of the prepositional progressive in Dutch (aan het-progressive), I claim that this approach is of added empirical and theoretical value. Drawing on construction-based theories, the relevant patterns—dubbed situational constructions—are analyzed in terms of horizontal constructional links.*

1. Introduction.

Research into progressive viewpoint aspect in Germanic usually proceeds in a top-down fashion. That is, the crosslinguistic concept progressive aspect serves as a vantage point from which one or more (cross)linguistic forms taken to be instantiations of it are described, analyzed, compared, and so on. Examples of progressive forms in the Germanic languages studied in this way include the English V-ing progressive (Ramchand 2018, chapter 2), the am-Progressiv in certain varieties of German (Krause 2002), pseudo-coordination in Norwegian, Swedish, and German (Tonne 2007, Blensenius 2015, Proske, this issue), the aan die-progressief in Afrikaans (Breed & van Huyssteen 2015, Wierenga 2022), and the Dutch aan het-progressief (Boogaart 1991; Lemmens 2005, 2015; Booij 2010, chapter 6). There are also studies that draw comparisons between progressives from these languages in various combinations (Boogaart 1999, chapter 5; Felser 2000; Van Pottelberge 2004, 2007; Behrens et al. 2013; Cavirani-Pots 2020; Wierenga & Breed 2021; Okabe 2023).

This is clearly a valid way of studying (progressive) aspect. However, this approach also harbors a certain risk: Examining a language-specific structure only in its capacity as an expression of one conceptual category may hide from view similar forms that do not instantiate the conceptual category under study. The risk is therefore that a form would be studied in isolation rather than in connection with related constructions, based on its place within a language-specific context. One way to avoid this problem—one that I advocate here—is to complement top-down treatments with a bottom-up approach. In the context of this study, such a bottom-up approach would specifically take into account structurally similar yet nonprogressive aspectual constructions.

Let me illustrate this point with the English V-ing progressive.¹ There is a vast body of work on present participles licensed by the progressive auxiliary be, as in Rosie was eating, but comparatively little on alternative aspectual auxiliaries used in this pattern, such as start, get, keep (on), stop, and finish, as in Rosie started/got/kept (on)/stopped/ finished eating, even though they, too, select present participles and—while not progressive—likewise encode the subject’s involvement in the situation denoted by the participle.²

To be sure, this does not invalidate studying be + V-ing as a dedicated progressive construction. However, such an approach does not address the question of how be V-ing relates to start/get/keep (on)/stop/ finish V-ing. Importantly, this type of question is not just empirically but also theoretically relevant against the backdrop of construction-based theories of language (see Goldberg 1995, Verhagen 2005, Hoffman & Trousdale 2013, Boogaart et al. 2014, Hilpert 2014, Diessel 2019, among many others), which model linguistic knowledge as a network of constructions, that is, symbolic pairings of form and meaning. A current debate in Construction Grammar (henceforth CxG; Goldberg 1995 and subsequent work) concerns the types of relations that exist among constructions within the constructional network. Specifically, it has been argued that so-called vertical relations (that is, inheritance links between constructions at different levels of abstraction)—which traditionally received most attention—are not sufficient to capture the generalizations that make up mental grammars, and that “horizontal” relations (that is, associations between constructions at the same abstraction level) are also required (for example, Van de Velde 2014, Audring 2019, Diessel 2019, chapter 10, Sommerer & Smirnova 2020). In the case of progressives in Germanic, defining one’s research object primarily or exclusively in terms of the top-down concept progressive may lead to empirical and theoretical gaps: Under this approach, the horizontal relations between these progressive constructions and similar (but not necessarily progressive in the traditional sense) patterns may be overlooked.³

To substantiate the general claim that a bottom-up approach is of added empirical and theoretical value to top-down studies of (progressive) aspect, this paper presents a case study conducted within the CxG framework. The study examines the Dutch pattern known as the aan het-progressive, or prepositional progressive (see, among others, Lemmens 2015 and Bogaards et al. 2022). This construction is illustrated in 1.⁴

1. (1)

   

The construction in 1 consists of the constituent aan het V_INF headed by the preposition aan ‘on’ combined with the matrix verb zijn ‘be’.⁵ It expresses that the situation denoted by the infinitive (here: zwemmen ‘swim’) is ongoing. This paper aims to show that this pattern shares crucial properties with certain other patterns built around aan-constituents, which differ from the aan het-progressive primarily with respect to the matrix verb or the variable element: The aan het-ingressives (Bogaards et al. 2022) in 2a combine with the matrix verbs gaan ‘go’ or slaan ‘hit’ rather than zijn, and the patterns in 2b,c select a variable verb stem and noun, respectively, rather than an infinitive.

1. (2)

   

Similarities in both the form and meaning of 1 and 2a–c complicate the idea of a dedicated, standalone or self-contained Dutch progressive. The main research question of this paper is therefore how the patterns exemplified by 1 and 2 are related from a constructionist point of view. Based on several semantic and syntactic similarities and contrasts, I argue that the aan het-progressive is a member of a broader category of prepositional aspectual constructions connected by horizontal (but not vertical) links.

The structure of the paper is as follows: Section 2 reviews previous work on aan het V_INF zijn as a progressive construction. In section 3, I lay down the analytical framework by discussing CxG and its concepts of constructional families and relations. Next, section 4 broadens the empirical scope of the study of aan het V_INF zijn by taking a bottom-up approach to prepositional aspectual constructions in Dutch. The observations made using this approach are discussed in section 5, which presents a novel analysis of these constructions in terms of their mutual links. Finally, section 6 concludes the paper.

2. The Aan Het-Progressive.

The Dutch construction known as the aan het-progressive, or the prepositional progressive (for example, Lemmens 2015, Bogaards et al. 2022) consists of a constituent aan het V_INF paired with the matrix verb zijn ‘be’. It presents the situation denoted by the variable infinitive as dynamic and ongoing, with the subject of zijn construed as the logical subject of the infinitive. An example of the construction is given in 3.⁷

1. (3)

   

In the introduction, I claimed that this construction has been dealt with mainly using a top-down approach to (progressive) aspect. By this I mean that it is a priori treated as a progressive construction and is analyzed or compared to other constructions only in that capacity. For example, Boogaart (1991) and Van Pottelberge (2007) contrast the aan het-progressive with the default nonprogressive form in Dutch, that is, the simple present/past. Lemmens (2005, 2015), Beekhuizen (2010, chapter 6), Behrens et al. (2013), and Anthonissen et al. (2019) compare this construction with other progressive forms, most notably the Dutch posture verb progressive, as in Rosie zit te eten ‘Rosie is eating’ lit. ‘Rosie sits to eat’. Finally, Boogaart (1999), Felser (2000), Van Pottelberge (2004, 2007), Behrens et al. (2013), and Wierenga & Breed (2021) compare it to progressive forms from other Germanic languages (namely, Afrikaans, English, German, Norwegian, and Swedish). Notable findings from this top-down-oriented work are that the Dutch aan het-progressive i) exhibits a high degree of grammaticalization compared to progressives in German, Norwegian, and Swedish (Van Pottelberge 2007, Behrens et al. 2013), ii) may denote a situation stretching over an extended period of time so that it covers multiple separate instantiations of that same situation (Vismans 1982, Lemmens 2015), iii) does not allow a habitual interpretation (Boogaart 1999), iv) may take on discourse-organizational functions (Boogaart 1991), and v) has lower (inter)subjective potential than the posture verb progressive (Anthonissen et al. 2019).

I should point out, however, that there is some notable work that goes beyond examining aan het V_INF zijn as an isolated progressive construction, which the present study takes as a starting point. For example, Bogaards et al. (2022) compare the aan het-progressive to the ingressive variants with gaan ‘go’ or slaan ‘hit’ as the matrix verb, as in 2a. They conclude that these are two different syntactic constructions and situate the aan het-progressive higher on a synchronic lexical-to-functional cline than the aan het-ingressive (that is, the former is more grammaticalized).⁸ Broekhuis et al. (2015:153) note the semantic similarity between the aan het-progressive and the verbal stem pattern in 2b.

Booij (2010:163–165) offers a construction-based analysis of both the aan het-progressive and its ingressive counterpart with slaan. He formalizes the connection between the aan het-progressive form and meaning as in 4a, and the ingressive with slaan as in 4b (from Booij 2010:164).

1. (4)

   

The schemata in 4 specify form-meaning links through coindexation of formal slots and sequences with semantic content (indices i-m in subscript). This notation captures the fact that each construction projects a different aspectual viewpoint (prog/ingr) onto the situation denoted by the infinitive (pred). Another well-known difference between these constructions is that the progressive but not the ingressive can optionally select an internal argument (see section 4.5). Booij (2010) accounts for this contrast by appealing to a general transitive schema [(argument) pred] in 4a, not present in 4b.⁹

In what follows, I build further on the accounts cited in this section by drawing on the concept of horizontal relations in CxG. In particular, building on Booij’s (2010) analysis, the present study examines how the formalizations in 4—as well as the examples in 2b,c—are related within the constructional network. Booij (2010:164) qualifies the aan het-ingressives in 4b as “subconstructions” of 4a, linking 4b to 4a vertically through instantiation. Note that one consequence of Booij’s approach is that ingressivity in 4b operates on progressivity, because it inherits the prog-semantics of 4a. This relation potentially introduces a conceptual problem, given that these types of aspect express diametrically opposed viewpoints: Progressivity defocuses a situation’s boundaries (Behrens et al. 2013:98), whereas ingressivity focuses its initial boundary (see Xiao & McEnery 2004, section 5.3). Hence [ingr[prog[pred]]] in 4b entails taking a situation and defocusing its boundaries, only to then (re)focus one of them—a somewhat roundabout procedure.¹⁰ In my view, this particular outcome of Booij’s analysis reflects the top-down approach that starts from the notion progressive (and the form that is assumed to be tied to this notion). The fact that Booij (2010) includes prog in the conceptual definition of an ingressive construction is characteristic of this type of approach. In section 4, I put forward an alternative to the relation between 4a and 4b as analyzed by Booij (2010).

3. Constructional Networks and Horizontal Links.

Construction-based approaches to language—most prominently CxG—model linguistic knowledge as symbolic pairings of form and meaning called constructions organized in a taxonomic network (sometimes referred to as “the Construction”; for example, Verhagen 2005:211, Van de Velde 2014:145). Constructions occupy a position in the network defined by their relations to other constructions (Diessel 2019:200). Together they form so-called construction families (ibid., 199). One way of organizing constructions is by vertical relations, from concrete to gradually more abstract linguistic patterns. To illustrate how this system works, Hilpert (2014:58) gives the example of the fully lexically specified idiom face the music with the noncompositional meaning ‘accept responsibility’. This idiom is one instantiation of the transitive verb face + OBJ, which in turn generalizes upward to the maximally schematic English transitive construction V OBJ. Vertical relations are thus token-to-type and type-to-token links, where token and type are relative terms: The expression face the music is a token that instantiates the type the transitive verb face + OBJ, which itself is a token that instantiates the type V OBJ.

As has been pointed out frequently in recent years, vertical relations are not sufficient to capture the associations between constructions (see, among others, Van de Velde 2014, Audring 2019, Diessel 2019, Sommerer & Smirnova 2020).¹¹ If there are similarities between constructions but no productive overarching construction they generalize to, these links cannot be modeled by appealing to abstraction. Instead, such constructions operate at the same level of schematicity without projecting upward to a more abstract (common) representation. Different terms are in circulation for this type of links between constructions, including “syntactic paradigms” (Van de Velde 2014), “sister relations” (Audring 2019), and “horizontal relations” (Van de Velde 2014, Diessel 2019, Sommerer & Smirnova 2020). As noted by Van de Velde (2014:141), horizontal relations “have been somewhat neglected in comparison with the vertical relations,” particularly at the syntactic level.

According to Diessel’s (2019:200) analysis of German, horizontal relations are defined by similarities and differences between constructions. As an illustrative example, he discusses copular clauses and stative passives in German, which are not linked vertically either to the same schema or to each other and yet exhibit considerable similarity. In Dutch, copular clauses and stative passives both consist of a subject, the copula/auxiliary zijn ‘be’, and a complement: an adjective in the former and a participle in the latter, that is, subj zijn adj/ptcp (I am using Diessel’s 2019:200 notation). So, although de lamp is aan ‘the lamp is on’ (copular) and de lamp is aangezet ‘the lamp has been turned on’ (passive) are instantiations of different constructions, they are alike formally (syntagms) and semantically (resultant states). Diessel thus argues that constructions such as subj zijn adj/ptcp are defined not just by vertical links to more general/specific schemata but also by mutual horizontal relations. These relations can be viewed in terms of the syntactic and semantic similarities and differences between the constructions in question. The next two sections analyze the aan het-progressive and similar patterns in these terms to establish the horizontal links between them.

4. A Bottom-Up Approach.

4.1. The Constructions.

In his work on the Dutch progressives, Lemmens (2005, 2015) terms the aan het-progressive—that is, aan het V_INF zijn in 5a—the “prepositional progressive”. This term highlights the fact that the preposition aan ‘on’ is the distinctive characteristic of the construction, as opposed to the Dutch posture verb progressive. If this language-specific form (rather than the top-down notion progressive) is taken as a starting point, it turns out that the construction also occurs with matrix verbs other than zijn—for instance, gaan ‘go’, as in 5b. Furthermore, the slot occupied by an infinitive can be filled by other forms too: a verbal stem, such as wandel ‘stroll’ in 5c (Ebeling 2006:112, Broekhuis et al. 2015:153, Booij & Audring 2018:220–223), or a noun, such as bier ‘beer’ in 5d (Boogaart 1999:169, Ebeling 2006:259, Lemmens 2015:8).¹²

1. (5)

   

The position I defend here is that Dutch has a constructional family of seemingly prepositional patterns, each of which encodes a particular type of viewpoint aspect. Starting out not from their aspectual meaning (the top-down approach) but from their form, one can see that these structures consist of the preposition aan ‘on’, a definite article, het/de, and a variable slot that can be filled by an infinitive, a verbal stem or a noun. The matrix verb in these constructions is optional (Van Pottelberge 2004:50, Bogaards et al. 2022). A schematic representation of these constructions is given in 6.

1. (6)

   

It is not evident a priori whether the constructions in 5 generalize vertically to a schema such as in 6; there is also another possibility, namely, that they are linked horizontally, as “a set of alternating forms with related meaning differences” (Van de Velde 2014:149). I argue that the latter is, indeed, the case.

4.2. Aan-Constructions Have Situational Rather Than Object Reference.

A crucial property shared by the examples in 5, regardless of the matrix verb, is their denotation: They all “receive a non-locational interpretation” (Booij 2010:153). In other words, they share situational reference. This nonlocational interpretation contrasts with the locational meaning of aan, namely, a spatial relation of contact between two entities, such that one “sticks to” the other at the area of contact (Beliën 2002:201), as illustrated in 7 (from Beliën 2002:207).

1. (7)

   

In 7, aan locates the painting on the wall. More generally, the complement of aan, de muur ‘the wall’, is construed as a locational landmark that refers to a specific object—a first-order entity (Lyons 1977:442–444) and a case of object-reference (Bierwisch 2011:336–338). By contrast, the examples in 5 encode not locations but situations.¹³ Their situational reference is derived from the element in slot X (for example, aan het bier ‘having a beer’ lit. ‘on the beer’). There is one difference in this respect between aan het V_INF and aan de V_STEM/aan D N: As Booij & Audring (2018) point out, the latter (but not the former; see Boogaart 1999:185) can also be interpreted habitually, as in aan het bier meaning ‘in the habit of drinking beer’.

In the case of an infinitive or a verb stem, the variable element itself has situational reference; but for nonprocess nouns such as bier ‘beer’, a key question is how situational reference is derived. Ebeling (2006:112) and Booij & Audring (2018:220) point out that this reference emerges through metonymic extension: aan DET N refers to a situation “in which the object denoted by the noun plays a central role” (Booij & Audring 2018:220). Under a metonymic interpretation, the element in slot X does not denote one particular object. Rather, this slot hosts second-order entities (Lyons 1977) with situational or event-reference (Bierwisch 2011): Aan het bier comes to denote an event of drinking beer, which may or may not be habitual. This secondary interpretation falls within Bierwisch’s definition of this type of reference as “entities that instantiate propositions and are subject to temporal identification” (2011:338). So rather than joining two first-order entities in a spatial relationship, situational aan introduces a second-order entity that can be predicated of an individual. Given that this property is shared by these particular phrases headed by aan, from now on I refer to this set of patterns as situational aan-constructions. Their situational denotation is contingent on several prerequisites, which are discussed next.

4.3. Aan-Constructions Do Not Exhibit Behavior Typical of PPs.

The previous section outlined a semantic distinction between regular locational PPs and the aan-constructions under investigation, which receive a situational interpretation. There are also at least two syntactic properties that set these situational patterns apart from locational ones. First, in locational PPs, the definite article may be replaced by any eligible determiner, and the noun may be modified. Taking 7 as an example, the definite article de in de muur ‘the wall’ can be substituted by an emphatic or an indefinite article or by a demonstrative, possessive, indefinite or distributive pronoun, as shown in 8a. Likewise, the noun may be modified adjectivally or prepositionally, as shown in 8b.

1. (8)

   

In contrast, the D and X slots in 6 are remarkably rigid: Unlike locational PPs, situational PPs do not allow the definite article to be replaced by any other determiner, and the variable element may not be modified. None of the modifications acceptable in 8 are possible in the situational constructions in 9.

1. (9)

   

The rigidity of the D and X slots is also observed in aan het V_INF/aan de V_STEM. It is thus this particular constructional family—consisting of aan het V_INF, aan de V_STEM and aan D N—that is tied to situational reference; if V_INF/V_STEM/N (that is, X) is modified or het/de (that is, D) is altered, the form-meaning pairing is lost. This similarity in behavior is expected if situational aan-constructions constitute a constructional family.

Second, situational aan-constructions diverge from regular PPs in not allowing r-extraction, that is, relativization of the complement of the preposition. R-extraction is typically associated with PPs in Dutch (see Broekhuis 2013:258–267 and references cited there), but it is not acceptable with situational aan-constructions, as demonstrated in 10. In 10a, de muur ‘the wall’ can be relativized with the R-pronoun waar ‘where’. In contrast, situational aan-constructions do not allow R-extraction, as shown in 10b.

1. (10)

   

These two differences between situational aan-constructions and locational PPs suggest that the former are not PPs, despite seemingly being built around a preposition (see Bogaards et al. 2022 for a proposal on the syntactic category of aan-constructions). At the same time, the similarities between aan het V_INF, aan de V_STEM, and aan D N support their status as a constructional family. However, there are also differences, which are discussed next.

4.4. Aan-Constructions Are Restricted to Certain Matrix Verbs.

Situational aan-constructions usually combine with some matrix verb, as shown in 5. When they do, they select verbs from the same limited pool, unlike locational PPs. The infinitival pattern aan het V_INF combines with a larger number of different verbs than aan de V_STEM/aan D N: There are 12 core verbs that are found with aan het V_INF. It has been observed that these verbs can be distinguished along two dimensions: i) progressive versus ingressive viewpoint, and ii) (non)causativity (Haeseryn et al. 1997:1048–1054; Van Pottelberge 2004:27–51; Booij 2010:146–168; Bogaards 2020:62–91, 2022:6–7; Boogaart & Bogaards 2023). Table 1 lists all 12 core verbs.¹⁴

Table 1. Matrix verbs that occur with aan het V_INF.

This observation is unidirectional: Locational aan-PPs are not barred from occurring with the verbs in table 1, but their situational counterparts are restricted to this verb pool. To illustrate, the locational PP from 7 may be combined with houden ‘keep’, krijgen ‘obtain’, and zijn ‘be’ from table 1, but also with other verbs, such as zich bevinden ‘be situated’, kleven ‘stick’, and prijken ‘adorn’, as shown in 11. Situational aan het bier ‘having beer’ in 12 is limited to the first three. Again, the combinatorial patterns in 12 apply to aan het V_INF and aan de V_STEM as well.

1. (11)

   

1. (12)

   

As observed before, not all matrix verbs in table 1 combine with aan de V_STEM/aan D N. For example, slaan ‘hit’ and maken ‘make’ in 13a and 13b, respectively, yield ungrammatical results.

1. (13)

   

To get a better empirical handle on the combinatorial tendencies of aan de V_STEM/aan D N with the matrix verbs in table 1, queries were conducted in the SoNaR corpus of written Dutch. A set of 10 common types of aan de V_STEM and aan D N (based on corpus data from SoNaR in Bogaards 2020) were selected, as shown in 14.

1. (14)

   

Tokens were then extracted with a finite verb directly preceding or following these aan-sequences. If the token in question behaved like a situational aan-construction (in the ways outlined in sections 4.2 and 4.3), then the finite verb was lemmatized. Table 2 shows the core matrix verbs that combine with the types of aan de V_STEM and aan D N in 14, ordered by absolute frequency. As expected, maken ‘make’ and slaan ‘hit’ are absent from table 2. Moreover, zetten ‘put’ does not combine with either pattern, and krijgen, komen, and houden are not attested with aan D N. Table 2 also contains one matrix verb that does select aan D N but is not present in table 1: the posture verb zitten ‘sit’. For aan D N, there seems to be free variation between zitten and zijn, as shown in 15.

Table 2. Matrix verbs that combine with each of the patterns in 14 (based on the data from SoNaR).

1. (15)

   

In sum, situational aan-constructions are only compatible with a limited number of matrix verbs, with some extra matrix verbs tied specifically to aan het V_INF and aan D N. Based on these data, there is not one generalization that captures the properties of the slot for the matrix verb in 6, as each of the three situational aan-constructions imposes its own restrictions.

4.5. Objects and High Particles Are Only Allowed with the Progressive.

The final property discussed here is a contrast between aan het V_INF on the one hand and aan de V_STEM/aan D N on the other. Bogaards et al. (2022) show that progressive aan het V_INF zijn and ingressive aan het V_INF gaan/slaan differ with respect to the possibility of direct objects and the placement of verbal particles (for example, op ‘up’ in opruimen ‘clean up’). This is demonstrated in 16 and 17 (from Bogaards et al. 2022:9). In 17a, opruimen cannot license the direct object z’n kamer ‘his room’, and in 17b, the verbal particle op can only be right next to ruimen.

1. (16)

   

1. (17)

   

The contrast between 16 and 17 leads Bogaards et al. (2022) to conclude that aan het V_INF zijn is a separate construction from aan het V_INF gaan/slaan; there are optional positions available for one object and one particle to the left of aan het—that is, (OBJ) (PTCL) aan het V_INF zijn—in contrast to the other type of aan het V_INF without those slots.

Looking at aan de V_STEM zijn/gaan, there is no such contrast: zijn and gaan behave the same with regard to objects and particles. To illustrate, 18a contains the particle verb opruimen. Sentence 18b shows that no matter which verb is used—gaan or zijn—there cannot be an object (m’n kamer ‘my room’) or high particle (*op aan de ruim). The construction aan D N patterns with aan de V_STEM in not exhibiting these contrasts because its variable element is not verbal.

1. (18)

   

There is now enough input for a constructional account of situational aan-constructions outlined in the next section.

5. A Family of Situational Aan-Constructions.

The previous section laid out relevant similarities and differences between aan het V_INF, aan de V_STEM, and aan D N. Starting out with the general meaning, these aan-constructions share situational reference. Borrowing Booij & Audring’s characterization of aan de V_STEM as “involved in the […] action [V_STEM]” (2018:223), the patterns’ denotation can be defined as in 19. For aan de V_STEM/aan D N, but not aan het V_INF, this involvement can be habitual. For aan D N, the derivation of situational reference from N is specified as metonymic.

1. (19)

   

The patterns in 19 differ from locational aan-PPs in several syntactic respects: The determiner het/de may not be substituted with another determiner, the variable element V_INF/V_STEM/N may not be modified, and the aan-constituent does not allow R-extraction. Moreover, the optional matrix verb V is restricted to a closed set, which varies according to construction. These matrix verbs contribute either a progressive or an ingressive aspectual viewpoint. The restrictions are listed in 20.

1. (20)

   

With aan het V_INF, the matrix verb determines whether there are positions for a direct object and a verbal particle to the left of aan. These positions are only allowed with zijn. Moreover, they are never available in the aan de V_STEM and aan D N constructions. Following Bogaards et al. (2022), I therefore assume that there are two syntactic constructions with an infinitival slot: one with zijn and one with another matrix verb, as shown in 21. The constructions aan de V_STEM and aan D N pattern with 21b regardless of their matrix verb. In aan het V_INF zijn on the one hand, and aan de V_STEM and aan D N on the other, the verb zijn thus does not have the same status: The verb zijn licenses the OBJ- and PTCL-slots in 21a in the former but not in the latter. Therefore, there is no single element V that would generalize over all three constructions. Property 20d must therefore be reformulated as 22d.

1. (21)

   

1. (22)

   

Taken together, the descriptions in 19–21 constitute a bundle of formal and functional properties defining a family of situational aan-constructions in Dutch, which the aan het-progressive is a member of. Based on these descriptions, the question arises as to how these constructions are related. Specifically, is there an overarching schema aan D X linking these constructions vertically (see 6), and/or are they connected horizontally?

The similarities in 20a–c suggest that, minimally, horizontal relations are needed to account for the properties shared by these constructions. Given the contrast between 21a, 21b, and 22d, I take the position here that an additional generalization along the lines of 6 is not warranted. For speakers to know which matrix verbs in 22d go with which patterns, these combinations must be stored separately, particularly considering the structural contrast in 21. These complex relations are better captured by horizontal links that would connect the shared matrix verbs in 22d, the semantics in 21, and the structural restrictions in 20a–c. Figure 1 visualizes the family of situational aan-constructions as conceptualized presently.¹⁵ Subscripts a–d (for example, d.1 in V_d.1) index the specific formal properties bundled in 20a–c and 22d. Subscript e indexes the viewpoint-aspectual semantics schematized in 20e.

Figure 1. A family of situational aan-constructions (indices refer to 20a–c,e and 22d).

Figure 1 postulates a larger family of aan-constructions than previously assumed (for example, in Booij 2010, Lemmens 2015), and, in my view, one that is also placed on a broader empirical footing—which is something gained by adding a bottom-up approach to the mostly top-down treatment of (progressive) aspect.

6. Conclusion.

This paper has advocated a bottom-up approach to constructions expressing (progressive) viewpoint aspect, in addition to more traditional, top-down treatments. In particular, I have drawn attention to the predominance of top-down approaches in the study of progressive constructions in Germanic and claimed that this could cause related (but not progressive) language-specific structures to be overlooked. The proposed approach is cast within construction-based theories of language, drawing on the concept of horizontal links.

To support my claim, I took a bottom-up approach to the Dutch aan het-progressive. The main insight gained from this case study is that this pattern is not an isolated aspectual construction but a member of a family of aspectual constructions all featuring the preposition aan ‘on’. All of the constructions have situational rather than locational meaning, and share a number of crucial formal properties as well. Based on several differences between these constructions (habituality, metonymy, matrix verbs, and positions for objects/particles), I argued that a vertical generalization over them is not warranted, and that they would be better accounted for using horizontal links. On a theoretical level, this case study provides evidence for the idea that horizontal links can be a better fit for the data than abstraction.

An appropriate closing observation for further research is that situational reference is not limited to the preposition aan ‘on’ in Dutch. There are myriad examples of other apparent PPs that receive this type of interpretation, including in de war zijn ‘be confused’ lit. ‘be in the tangle’, op hol slaan ‘run wild’ lit. ‘hit on run’, van start gaan ‘commence’ lit. ‘go from start’, and many more. I hope to have argued convincingly that a relevant next step would be to extend the approach developed here to these patterns—and to (progressive) aspectual constructions in other (Germanic) languages.